Using LSTM Neural Network Based on Improved PSO and Attention Mechanism for Predicting the Effluent COD in a Wastewater Treatment Plant

Abstract

Enhancing the monitoring capabilities of wastewater treatment plant (WWTP) key features can accomplish accurate prediction to help WWTPs develop a plan, which is of great significance to control regional water environmental pollution. Chemical oxygen demand (COD) is one of the key features of wastewater treatment. Traditional monitoring methods are time consuming and have high costs making it difficult to meet the needs of rapid monitoring in practical applications. To address this issue, a method for optimizing a long short term memory (LSTM) neural network model based on adaptive hybrid mutation particle swarm optimization (AHMPSO) and an attention mechanism (AM) is proposed. As the hyperparameters of the LSTM are difficult to select, AHMPSO is used to optimize the LSTM. A nonlinear variable inertia weight with random factors is introduced to balance the global search ability and the local search ability and to improve the convergence speed of the PSO algorithm. In addition, the hybrid mutation strategy is added in the search process to reduce the risk of particles falling into local optimal solutions. Finally, an AM is added to the LSTM model to mine local water quality features to improve the effluent COD prediction accuracy. Compared with other models (LSTM, LSTM-AM, and PSO-LSTM-AM), the RMSE(the root mean square error) of the optimized model decreased by 7.803%-19.499%, the MAE(the mean absolute error) of the optimized model decreased by 9.669%-27.551%, the MAPE(the mean absolute error) of the optimized model decreased by 8.993%-25.996%, and the R2 (the coefficient of determination) value of the optimized model increased by 3.313%-11.229%. The experimental results show that the optimized model has better performance and achieves a more accurate prediction of the effluent COD.